NYU Tandon’s new AI system can detect fires within 0.016 seconds using standard security cameras, offering a groundbreaking solution for faster evacuation and emergency response times.
Researchers at the NYU Tandon School of Engineering have developed a revolutionary artificial intelligence system that can detect fires almost instantaneously using standard security cameras. This innovation promises to significantly enhance fire safety, potentially saving lives and reducing property damage.
Fire-related incidents claim nearly 3,700 lives annually in the United States and cause approximately $23 billion in property damage. Traditional smoke detectors often fail to provide timely alerts, especially in large, open spaces.
The NYU Tandon system aims to bridge this gap by identifying fires and smoke in real-time, providing crucial extra minutes for evacuation and emergency services.
Published in the IEEE Internet of Things Journal, the study highlights that the AI system can analyze video footage and identify fires within 0.016 seconds per frame. This speed, faster than the blink of an eye, could revolutionize emergency responses.
Lead researcher Prabodh Panindre, a research associate professor in NYU Tandon’s Department of Mechanical and Aerospace Engineering, emphasized the system’s wide-reaching potential.
“The key advantage is speed and coverage,” Panindre said in a news release. “A single camera can monitor a much larger area than traditional detectors, and we can spot fires in the initial stages before they generate enough smoke to trigger conventional systems.”
The need for improved fire detection technology is urgent. Statistics reveal that 11% of residential fire fatalities occur in homes where smoke detectors either failed to alert occupants or were nonexistent. Moreover, modern building materials and open floor plans make fires spread quicker and reduce structural collapse times compared to older constructions.
To address these challenges, the NYU Tandon team used an ensemble approach, combining multiple advanced AI algorithms. By requiring agreement among these algorithms, the system significantly reduces false alarms — a critical factor in emergency situations.
The researchers created a comprehensive image dataset to train their models, representing all five fire classes recognized by the National Fire Protection Association. The system achieved an impressive 80.6% detection accuracy, with the best-performing model combination.
Additionally, the team incorporated temporal analysis to differentiate between real fires and static fire-like objects, further minimizing false alarms.
“Real fires are dynamic, growing and changing shape,” added Sunil Kumar, a professor of mechanical and aerospace engineering. “Our system tracks these changes over time, achieving 92.6% accuracy in eliminating false detections.”
This cloud-based system utilizes existing CCTV infrastructure, performing AI analysis on servers where video is streamed. When fire is detected, the system sends real-time alerts via email and text message, making it cost-effective for widespread adoption.
Beyond urban settings, this technology can be integrated into drones or unmanned aerial vehicles for early-stage wildfire detection, potentially saving critical hours in extinguishing fires and prioritizing evacuation orders.
To assist firefighters, the system can also be embedded into tools they already use, such as helmet cameras, thermal imagers and vehicle-mounted cameras. In urban firefighting, drones equipped with this AI could provide 360-degree fire size-ups, especially in high-rise buildings.
“It can remotely assist us in confirming the location of the fire and the possibility of trapped occupants,” added Capt. John Ceriello of the Fire Department of New York City.
The research team sees potential for their approach to extend beyond fire detection. Their system could be adapted to monitor and respond to other emergencies, such as security threats or medical crises, broadening its applicability and impact on public safety.
Source: NYU Tandon School of Engineering